The shared vertex/index arena is the largest single GPU allocation acdream
makes (384 MiB + 128 MiB) and the one the Vulkan backend has the most specific
plan for (campaign doc section 4.3). This slice swaps the resource handle type
underneath it and changes nothing else: the reclaimable-range allocator, the
growth quanta, the budgeted incremental grow-and-copy, the retirement-ledger
gating, the abort ticket, the LRU that drives eviction, and the 896 MiB
dual-generation physical ceiling are all untouched. That is deliberate - those
are the semantics section 4.3 says the Vulkan arena must mirror exactly, so
preserving them is the point of the slice rather than an incidental constraint.
What moved:
- GlobalMeshBuffer's two GL buffer objects became IGpuBuffer, allocated through
IGpuDevice.CreateBuffer with DeviceLocal residency and Vertex-or-Index plus
both transfer usages (the arena is simultaneously a draw source and both ends
of its own migration, which is exactly why GpuBufferUsage is a flags enum).
- UploadMesh's two hand-rolled BufferSubData sites became IGpuBuffer.Upload.
The old code staged indices through GL_COPY_WRITE_BUFFER specifically so an
upload could not mutate whichever VAO a preceding render pass left bound;
Upload stages through a neutral binding point of the backend's choosing, so
that property now comes for free instead of by hand.
- AdvanceMigration's CopyBufferSubData became IGpuBuffer.CopyTo - a device-side
copy, which the Vulkan backend will record as vkCmdCopyBuffer. The live
prefix still never round-trips through system memory.
- BeginMigration/CommitMigration/AbortMigration/Dispose now carry IGpuBuffer in
the migration record and the abort ticket instead of raw uint names, so the
ticket's identity check is a resource identity rather than a number that goes
stale the moment the buffer is deleted.
What deliberately did not move. A VAO has no RHI verb - Vulkan bakes vertex
input into the pipeline - and WbDrawDispatcher, EnvCellRenderer and
ParticleRenderer still bind VAO/VBO/IBO with raw GL until V4c hands them the
pass encoder. So GlobalMeshBuffer keeps its GL handle for the vertex array and
its attribute layout, and VBO/IBO became computed properties that publish the
backing GL name of the buffer the arena now owns as an IGpuBuffer. One private
RequireGlBuffer helper is the single place that reaches through the interface,
and it disappears with those consumers. ObjectMeshManager therefore needed no
upload-path change at all - it reads those same three properties.
Two decisions worth recording.
First, arena deletes do not route through IGpuBuffer.Dispose. The arena already
gates every delete behind its own GpuRetirementLedger and decrements its
physical-capacity accounting in the same retirement stage; Dispose would defer
the physical free through the device queue a second time, so the accounting
would run ahead of real GPU residency and could admit a migration that breaches
the 896 MiB ceiling. GlGpuBuffer gains DeleteRetired for callers that have
already proved flight safety, and GlobalMeshBuffer composes it into a release
whose four stages match TrackedGlResource.CreateRetryableBufferDeletion exactly
- precondition, mutation-with-validation, byte accounting, resource-count
accounting - so a driver failure re-issues only the delete and never
double-counts.
Second, two corrections in the GL backend, both required to keep this port
behaviour-preserving rather than merely compiling. GlGpuBuffer's glBufferData
usage hint now follows residency (DeviceLocal -> StaticDraw), which is what the
arena has always requested; the host-writable rings and texture table keep
DynamicDraw and are unaffected. And a failed allocation now releases the GL
name it had already created - GL_OUT_OF_MEMORY is a real outcome for a 384 MiB
growth destination, and the previous code leaked the name on that path.
Plumbing: the device reaches the arena through WbMeshAdapter and
ObjectMeshManager. Their constructors became internal because IGpuDevice is an
internal type by the pinned contract, matching what V4a did for BitmapFont,
DebugLineRenderer and TextRenderer; both classes stay public and every caller
already lives inside AcDream.App or its InternalsVisibleTo test assemblies. The
unused public GlobalMeshBuffer(GL) convenience constructor is gone - it could
not supply a device and had no callers.
Gates. Release build green with TreatWarningsAsErrors. App tests 3,843 passed /
3 skipped, exactly the slice baseline; complete Release suite 8,906 passed / 5
skipped. Offline pixel gate against 79ee2361: 25 differing pixels of 563,200
(fraction 4.44e-05), against a same-commit control captured immediately
afterwards of 24 - the change is indistinguishable from capture noise and sits
40x under the 0.001 threshold. An earlier gate run was discarded rather than
interpreted: its client log showed real ScrollUp/ScrollDown input reaching the
offline window, which zoomed the camera, and a camera-motion difference is not
a rendering result.
No divergence-register row: this slice changes no retail-facing behaviour.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Moves the mesh/EnvCell draw path's per-batch texture representation from a
64-bit ARB_bindless_texture handle to a small integer table index, entirely
on the still-shipping GL backend, with zero pixel change. This is the CPU-side
half of the eventual Vulkan descriptor-array indexing model: a table index is
the backend-neutral form (Vulkan indexes a descriptor array with it directly),
while a raw bindless handle is GL-only. Landing the data-model change now, on
GL, under a strict self-differential pixel gate, keeps it separate from V4c's
much larger RHI-plumbing change (see docs/plans/2026-07-27-vulkan-campaign.md
section 5.2 for why the table cannot be device-owned yet).
Mechanism: mesh_modern.vert's BatchData struct carries `textureIndex` (a slot)
instead of `textureHandle` (uvec2); the vertex shader looks the slot up in a
new binding=9 storage buffer (GpuBindingModel.StorageTextureTable) and passes
the reconstructed uvec2 handle to the fragment shader exactly as before, so
mesh_modern.frag needed no change at all beyond the UBO-set macro below. The
16-byte std430 stride is unchanged (GpuBindingModel.GpuBatchDataStrideBytes);
textureLayer/flags keep their offsets, so every existing CPU writer's layout
is untouched.
The handle->slot table (GlBindlessHandleTable, new, pure C#) is owned
separately by WbDrawDispatcher and EnvCellRenderer rather than shared through
a single TextureCache-owned instance: EnvCellRenderer never had a TextureCache
dependency, and nothing requires index agreement between renderers since each
rebinds its own binding=9 buffer immediately before its own draw call. This
avoided threading a new constructor parameter through EnvCellRenderer (and its
six test call sites) for no behavioral benefit. TextureCache and
CompositeTextureArrayCache turned out to need no changes at all: they only
ever produce raw ulong handles, and that production path is unaffected -
the new indirection is entirely a WbDrawDispatcher/EnvCellRenderer-side
concern, added exactly where each already assembles its per-batch GPU struct
(ToInput, the copy-back loop, PrepareDeferredAlphaDraws for the
RetailAlphaQueue path, and EnvCellRenderer's ModernBatchData construction).
The table itself is a single non-ring buffer (unlike the per-frame
triple-buffered SSBOs) because a genuinely new handle is rare - new dat
surfaces/composite overrides, not every frame - so it flushes only when
GlBindlessHandleTable.Dirty is set, mirroring how the existing texture caches
already upload infrequently.
Shader-side, introduced Rendering/Shaders/common.glsl as the shared preamble
GL has no #include for: Shader.cs gained an `includeCommonPreamble` overload
that splices the file's text in after the leading #version/#extension block
(GLSL requires #version first). It declares the binding=9 table plus the
ACDREAM_TEXTURE_HANDLE(idx) lookup macro, and a scaffolding ACDREAM_UBO_SET
macro (a no-op under GL today, redefined to `set = 1,` when the Vulkan
toolchain compiles this same source at V6+, per the campaign doc's set-1 UBO
note) applied to both SceneLighting UBO declarations now so no later slice
needs to touch them again.
Tests: WbDrawDispatcherIndirectBuilderTests updated for the renamed
IndirectGroupInput/BatchDataPublic fields; new ModernBatchDataLayoutTests
(mirrors ClipFrameLayoutTests' role, but for EnvCellRenderer's GPU struct) and
GlBindlessHandleTableTests (pure-CPU allocator behavior, including the
zero-handle case, which is registered like any other handle rather than
special-cased, since that's what reproduces the pre-V2 sampling result
bit-for-bit).
Gate: dotnet build -c Release green, dotnet test
tests/AcDream.App.Tests -c Release green (3843 passed / 3 skipped, +9 over
the 3834/3 baseline), and tools/run-offline-pixel-gate.ps1 passed with a
2.84e-05 differing-pixel fraction against the parent commit - within the
documented ~33x same-commit noise margin. No divergence-register row: this
introduces no retail behavior deviation.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Move the sole PhysicsEngine, production cache, collision admissions, canonical bodies and hosts, remote components, ordinary/remote worksets, simulation, cell commits, and shadow synchronization under RuntimeEntityObjectLifetime. Keep App as the prepared-asset, animation-input, and render-projection adapter while preserving the named-retail update and collision order.
Add exact-incarnation, object-clock, callback-reentrancy, GUID-reuse, two-runtime isolation, source ownership, collision publication, and graphical projection coverage. Release build and the complete 8,588-test solution pass.
Co-authored-by: Codex <noreply@openai.com>
Move the canonical local movement controller, body/motion managers, object clock, movement wire data, and MTS/jump/AP sender into AcDream.Runtime. Replace process skill defaults with typed Runtime character options, make graphical and direct commands borrow one autorun owner, retain the construction-time PartArray seam, and include movement in terminal ownership convergence.
Preserve the accepted pre-inbound movement/jump and post-inbound autonomous-position order while moving the exact packet/cadence fixtures into Runtime tests. Add graphical/direct parity, two-instance isolation, teardown, allocation, architecture, and divergence-path coverage.
Co-authored-by: Codex <noreply@openai.com>
Unify the toolbar shortcut manager with Runtime inventory state, route retail-ordered shortcut and spellbook command effects through the canonical owners, and make retained controllers borrow those exact instances. Remove the item-interaction transaction fallback and add graphical/no-window parity plus failure-safe terminal ownership-ledger coverage.
Co-authored-by: Codex <codex@openai.com>
Move the retail one-request-at-a-time gate, shared use busy references, external-container state, item mana, shortcuts, and desired-component snapshots into one Runtime-owned graph over J3's exact ClientObjectTable. Retained UI and session routing now borrow that owner; reset and shutdown preserve the existing order while failure/reentrancy tests protect the transaction boundary.
Co-authored-by: Codex <codex@openai.com>
Publish prepared GfxObj, Setup, CellStruct, and EnvCell collision records without retaining their parsed DAT BSP, polygon, vertex, or shape graphs. Strip temporary physics bundles at stable world commit, keep graph traversal only as an explicit test/tooling oracle, and report graph residency from actual retained fields.
Validated by 115 focused collision/streaming tests, a zero-warning Release build, and 8,413 passing Release tests with five pre-existing skips.
Co-authored-by: Codex <codex@openai.com>
Eliminate boxed production surface-override enumeration, retain vital modifier projections until the enchantment registry mutates, and measure DAT font widths without allocating a captured delegate. Preserve exact hashes, spell stacking, and glyph advances with warmed zero-allocation tests.
Validated by the focused rendering, UI, and spell suites, a zero-error Release build, and 8,409 passing Release tests with five pre-existing skips.
Co-authored-by: Codex <codex@openai.com>
Make prepared flat BSP data authoritative for gameplay while retaining the parsed graph only as an exact sampled referee. Fail production publication when collision package data is genuinely absent, keep idempotent already-cached publication valid, and move cell membership, floor lookup, camera diagnostics, and live/static shape bounds onto the flat representation.
Validated by 8,402 Release tests, a strict dense-Arwic connected gate with 46,309/46,309 exact referee matches, and graceful shutdown.
Co-authored-by: Codex <codex@openai.com>
Carry one immutable prepared collision closure with each accepted near-tier generation and install graph plus flat views through the same retained publication receipt. Apply the same strict package-only rule to live entities, add exact sampled graph-authoritative comparison artifacts and lifecycle counters, and prove cancellation, demotion, rehydrate, revisit, teardown, reconnect, and the nine-stop route with 14,064 zero-mismatch samples.
Co-authored-by: OpenAI Codex <codex@openai.com>
Port every current containment, overlap, walkable, and six-path moving-collision query to immutable integer-indexed assets behind a graph-authoritative referee. Exact synthetic, installed-DAT, complete-resolver, and zero-allocation gates prove bit-identical behavior before connected dual publication.
Co-authored-by: OpenAI Codex <codex@openai.com>
Append strict collision/topology payloads to the existing prepared package so later physics cutover can drop parsed DAT graphs without adding a second mapping or changing traversal behavior. The full 2,232,170-key catalog is deterministic across worker counts, exact-byte aliased, corruption-isolated, and cancellation-safe.
Add immutable indexed physics and containment BSP records, exact-bit polygon and Setup payloads, separated CellStruct/topology ownership, and iterative positive-before-negative flattening. Reject malformed graphs and ranges, and prove source identity over synthetic edge cases and installed retail DAT samples without cutting production traversal over.
Mirror retail's ten-deep LIFO transition lifetime, retain all query scratch with complete reset contracts, and remove Tier-0 enum boxing without changing collision decisions. Fresh and retained engines are bit-identical across the expanded oracle, while measured transition profiles now allocate 0 bytes per resolve.
Detach old-world spatial ownership atomically, prioritize destination retirement dependencies, and reveal the viewport at the retail transition edge. Give private paperdoll views independent mesh ownership and retain dormant ACE entities so portal revisits preserve server objects without extending active GPU lifetimes.
Replace over-cap full sorting with a retained exact top-k heap while preserving the accepted tie-order fallback. Differential tests lock randomized and Town Network-scale output, and the measured 463-light path cuts selector CPU by 29 percent without warmed allocations.
Join destination scheduling to the canonical reveal generation, protect its share across every typed frame-budget dimension, and prevent stale work from clearing a replacement reservation. Remove forced incomplete materialization and project retail's centered portal wait cue while the authored tunnel remains active.
Tests: Release build clean; 91 focused reservation/reveal tests; full solution 8,158 passed, 5 skipped.
Co-authored-by: Codex <noreply@openai.com>
Publish the retail blocking-for-cells edge before deferred recenter work, freeze old-world presentation/simulation/audio, and advance full-window retirement from exact metered entity and owner cursors. This removes synchronous portal teardown without allowing retained owners to remain observable.
Complete Slice D4 by adding aggregate lifecycle occupancy and traffic facts, validating physical source reports, and including decoded audio under the typed startup budget. Exercise every domain under forced pressure and retain the real cache/fence convergence gates.
Two audit-verified caches grew monotonically for process lifetime, which is
fatal for the 30-bot long-uptime headless-fleet goal:
- DatSoundCache._waves (Core) memoized every decoded PCM WaveData forever
in a bare ConcurrentDictionary — no LRU, no byte budget.
- OpenAlAudioEngine._bufferByWaveId (App) retained a native OpenAL buffer
copy of the same PCM per wave id until engine disposal — a second,
independent unbounded cache.
DatSoundCache now bounds payload residency with a 32 MiB byte-budget LRU
(decoded PCM waves run ~100-500 KB each, so this holds a comfortable
working set). Missing/unsupported-format waves are memoized separately in
an unbounded-but-cheap negative-result set (bounded by the finite Wave dat
id space) so they can never compete with or get evicted alongside payload
entries. Concurrent first-touch decodes of the same wave id are deduped
via a shared Lazy<T> so racing callers don't pay for WaveDecoder.Decode
twice. AcDream.Core cannot reference AcDream.Content (code-structure rule
2), so the LRU is a local reimplementation mirroring
BoundedDatObjectCache/DecodedTextureCache's shape rather than a shared
dependency.
OpenAlAudioEngine._bufferByWaveId now bounds native buffer residency with
a 48 MiB byte-budget LRU (AlBufferBudgetTracker), evicting least-recently-
used buffers once oversized. alDeleteBuffers fails on a buffer still
attached to a source, so eviction queries live AL per-source state
(GetSourceInteger.Buffer) rather than tracking a second, easily-stale
copy — several call sites (Play3DWave, PlayUiWave) set a source's buffer
directly. The buffer EnsureBuffer just created is explicitly protected
from its own eviction pass, since the caller hasn't attached it to a
source yet at that point. Evicted waves simply replay through
DatSoundCache -> EnsureBuffer on next use, identical to a first play.
Verified before implementing: AudioHookSink is the only GetWave caller
(single per-frame render-thread path per AnimationHookRouter's own
threading doc), and PcmBytes is read only at DatSoundCache.Admit (byte
accounting) and EnsureBuffer's first-upload branch — confirmed dead after
AL upload on the steady-state replay path, so bounding either cache
independently is correctness-safe; a cold replay after both evict simply
falls back to a full re-decode + re-upload, identical to a first play.
AlBufferBudgetTracker's eviction/budget decision is extracted as pure
logic (no AL dependency) specifically so it's unit-testable: the existing
OpenAlResourceLifetimeTests fake exposes a null AL, which short-circuits
every native buffer call before it runs, so the engine's actual AL wiring
isn't testable headless.
Tests: 9 new DatSoundCacheTests (Core.Tests) covering eviction order, byte
accounting, negative-result memoization, oversize-single-entry handling,
and concurrent-access smoke tests; 10 new AlBufferBudgetTrackerTests
(App.Tests) covering the pure LRU/budget/protection logic. Full suite:
3214/2 skip (Core.Tests), 3471/3 skip (App.Tests) plus one pre-existing,
unrelated failure (LandblockBuildOriginTests.FarLoad_..., reproduces
identically with these changes stashed out — landblock streaming, not
audio).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(cherry picked from commit 76c880d35bcf30b50e3f7b4cb8635dc9ab9e3ec7)
Carry PublicWeenieDesc material type into the live object model so examination titles use the DAT-authored material prefix. Preserve retail AddItemInfo empty appends and embedded armor separator, restoring the deliberate blank rows between appraisal sections.
Co-authored-by: Codex <codex@openai.com>
Preserve PublicWeenieDesc hook identity from CreateObject through the item model so hook appraisals suppress sentinel capacities exactly. Use appraisal-only Value and Burden presence, retain AddItemInfo paragraph and authored font-color selection, and port retail lock, page, enchantment, and spell-block formatting.
Co-authored-by: Codex <codex@openai.com>
Preserve public shared-cooldown metadata, resolve the authoritative cooldown enchantment with retail expiry semantics, and project the exact ten DAT-authored radial steps through the shared retained item-slot architecture.
Co-authored-by: Codex <codex@openai.com>
Use the exact low USEABLE_NO-bit predicate so reset/zero-valued direct-use items such as Blackmoor's Favor reach the ordinary Use request. Port AutoWear's clothing-priority blocker lookup and exact named system notice through the shared activation owner.
Co-authored-by: Codex <codex@openai.com>
Port the retail selected-object availability predicate into Core and project it through the shared interaction owner. The imported hand now follows canonical selection/object notices, keeps weapon and targeted-tool activation on the existing wield/use cursor paths, and ghosts empty or explicitly unusable selections per the connected UX requirement.
Co-authored-by: Codex <codex@openai.com>
Preserve the fixed per-cell collision walk across nested registry mutations while replacing the repeated List allocation with an explicitly owned pooled snapshot. Capture cardinality once, return storage on every exit, and pin live-list mutation behavior with a focused regression test.
Co-authored-by: Codex <codex@openai.com>
Move camera pointer, framebuffer resize, and retained/devtools input edges behind focused reversible owners. Preserve input priority while making shutdown deactivate callbacks before live-session retirement and retry physical detach without stranding transport teardown.
Move local teleport, player-mode, animation, shadow, and sealed-dungeon lifetimes out of GameWindow. Make player-mode entry transactional and isolate approach completions by controller lifetime and approach generation so stale callbacks cannot affect replacements.
Co-authored-by: Codex <noreply@openai.com>
Move streamed terrain/cell/building and static collision publication behind a focused update-thread owner. Preserve retail publication order while making replacement and retirement exact by logical landblock ownership, including current-cell rebasing and adjacent-seam isolation.
Co-authored-by: Codex <noreply@openai.com>
Bind queued actions and pending inventory requests to exact live incarnations, separate optimistic placement from authoritative responses, and serialize retail-style inventory ownership across UI surfaces.
Co-authored-by: OpenAI Codex <codex@openai.com>
Carry local WorldEntity identity through render hits, lighting pulses, and deferred movement actions so GUID reuse cannot target a replacement. Reset all session-owned selection and ItemHolder state and prevent combat auto-target during teardown.
Make initial login and portal arrival consume one WorldRevealReadinessBarrier that joins near-tier mesh publication, destination composite uploads, and collision residency before normal world geometry becomes visible. This ports retail SmartBox's blocking-cell completion edge into the asynchronous client instead of exposing a ground-only login.
Add focused outdoor, indoor, texture, and invalid-claim tests; update the retail pseudocode, architecture, divergence record, issue ledger, roadmap baseline, and synchronized agent guidance.
Co-authored-by: OpenAI Codex <codex@openai.com>
